Small Block Audio Codec

Small Block Audio Compression
* Goal is for a simple hardware-decodable audio codec.
* Should allow random access to compressed audio samples.
** Random access disallows any dependency on prior sample values.


Small Block 1:
* Encodes 16 PCM samples into 32 bits (2 bits/sample).
* ( 5: 0): Line Start (S.E3.F2)
* (11: 6): Line End   (S.E3.F2)
* (15:12): Line Sigma (  E3.F1)
* (31:16): Sample Selector Bits

To decode a sample, linearly interpolate between the start and end value. The start and end values represent a line segment.

The Sigma will represent a distance above or below this line segment, which will be either added or subtracted from the interpolated value (based on the selector bit) to get the result.

For each selector Bit:
* 0: V=Pred-Sigma
* 1: V=Pred+Sigma


Small Block 2:
* Encodes 16 PCM samples into 64 bits (4 bits/sample).
* ( 7: 0): Line Start (S.E3.F4)
* (15: 8): Line End   (S.E3.F4)
* (23:16): Line Sigma (Z.E3.F4)
* (31:24): Reserved (Zero)
* (63:32): Sample Selector Bits

This is based on a similar premise for Format 1, just with higher endpoint fidelity.

While one "could" make the endpoint values bigger here, this doesn't actually gain much in terms of quality.

The interpolation values will also be 2-bits / sample:
* 00: Small Negative (-0.333 * Sigma)
* 01: Large Negative (-1.000 * Sigma)
* 10: Small Positive (+0.333 * Sigma)
* 11: Large Positive (+1.000 * Sigma)


Endpoints are encoded as a microfloat value which expands into a 12-bit integer format (PCM):
0: 00000000xxxx
1: 00000001xxxx
2: 0000001xxxx0
3: 000001xxxx00
4: 00001xxxx000
5: 0001xxxx0000
6: 001xxxx00000
7: 01xxxx000000

Negative values will invert the value of the decoded sample.
The smaller variants are similar, just with fewer fractional bits.

Interpolation and other calculations will be performed in terms of PCM values.
An encoder should ensure that decoded values will not go out of range.